New
mll-lab-nu/embodied-first-exam — a code-agent exam built on fixed scenes
Introduces an exam system in which a code agent solves robotic tasks drawn from several benchmarks as programs and is then verified in a separate sandbox.
Introduces an exam system in which a code agent solves robotic tasks drawn from several benchmarks as programs and is then verified in a separate sandbox.
View on GitHub ↗Introduces an exam system in which a code agent solves robotic tasks drawn from several benchmarks as programs and is then verified in a separate sandbox.
No matching stories.